Statistical Utterance Selection Using Word Co-occurrence for a Dialogue Agent
In this paper, we proposed a statistical utterance selection method for dialogue agents by applying a machine learning algorithm. We defined statistical candidate utterance selection as a question that automatically selects an appropriate utterance from speech collections prepared in advance as responses. To realize automatic utterance evaluation, we employed manually evaluated data as learning data so that relative magnitude correlation will be learned from them.
We checked the order of the automatically evaluated values to prove the validity of our proposed method. In this simulation, the result shows that the top appropriate utterance is selected at 47.5%, and it is selected within the top 10 at 78.0%. For implementing this method in agents that assist humans by replying, we found that it is quite possible to realize such an agent.
Keywordsmachine learning utterance selection dialogue agent non-task-oriented dialogue
Unable to display preview. Download preview PDF.
- 2.Dunning, T.: Accurate methods for the statistics of surprise and coincidence. Computational linguistics 19(1), 61–74 (1993)Google Scholar
- 3.Haruno, M., Shirai, S., Ooyama, Y.: A japanese dependency parser based on a decision tree. Transactions of Information Processing Society of Japan 39(12), 3177–3186 (1998)Google Scholar
- 4.Inaba, M., Isomura, N., Toriumi, F., Ishii, K.: Evaluation method of Non-task-oriented dialogue system using Semantic Network. Technical report of IEICE (2009)Google Scholar
- 5.Isomura, N., Toriumi, F., Ishii, K.: Evaluation method of non-task-oriented dialogue system using HMM. The IEICE transactions on information and systems J92-D(4), 542–551 (2009)Google Scholar
- 6.Kudo, T., Kazawa, H.: Web japanese n-gram version 1, published by gengo shigen kyokaiGoogle Scholar
- 7.Kudo, T., Matsumoto, Y.: Japanese dependency parsing using relative preference of dependency. Transactions of Information Processing Society of Japan 46(4), 1082–1092 (2005)Google Scholar
- 10.Uchimoto, K., Sekine, S., Isahara, H.: Japanese dependency structure analysis based on maximum entropy models. Transactions of Information Processing Society of Japan 40(9), 3397–3407 (1999)Google Scholar